feat(open-swe): Default to GPT-5.5 medium reasoning (#1224)

* feat: default to GPT-5.5 medium reasoning

Use OpenAI GPT-5.5 with medium reasoning as the default model and document the completion-token budget semantics for reasoning models.

* fix: use Responses API reasoning config

Pass GPT-5.5 reasoning settings through LangChain's Responses API parameter instead of the Chat Completions-only reasoning_effort field.

* feat: raise GPT-5.5 output budget

Set the default GPT-5.5 output token budget to the model maximum so long-running coding tasks have more room for reasoning and final responses.

* feat: align recursion limit with Deep Agents

Use Deep Agents' default recursion limit so longer coding runs have room to complete without Open SWE imposing a lower cap.

* chore: remove minimal effort level

* chore: reduce max tokens to 64_000

---------

Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com>

.gitignore

@@ -24,6 +24,11 @@
 # misc
 .DS_Store
 *.pem
+*.key
+*.crt
+*.p12
+*.pfx
+*.jks
 
 # debug
 npm-debug.log*
@@ -33,6 +38,7 @@ yarn-error.log*
 # local env files
 .env*.local
 .env
+.env.*
 
 # vercel
 .vercel
@@ -52,6 +58,7 @@ credentials.json
 apps/cli/test_traces/
 
 # Python
+.venv/
 __pycache__/
 **/__pycache__/
 *.py[cod]

CUSTOMIZATION.md

@@ -4,8 +4,13 @@ Open SWE is designed to be forked and customized for your org. The core agent is
 
 ```python
 # agent/server.py — the key lines
+model_id = os.environ.get("LLM_MODEL_ID", DEFAULT_LLM_MODEL_ID)
+model_kwargs = {"max_tokens": DEFAULT_LLM_MAX_TOKENS}
+if model_id == DEFAULT_LLM_MODEL_ID:
+    model_kwargs["reasoning"] = DEFAULT_LLM_REASONING
+
 return create_deep_agent(
-    model=make_model(os.environ.get("LLM_MODEL_ID", DEFAULT_LLM_MODEL_ID), temperature=0, max_tokens=20_000),
+    model=make_model(model_id, **model_kwargs),
     system_prompt=construct_system_prompt(...),
     tools=[http_request, fetch_url, list_repos, get_branch_name, commit_and_open_pr, linear_comment, slack_thread_reply],
     backend=sandbox_backend,
@@ -119,14 +124,16 @@ See `deepagents.backends.LangSmithSandbox` and `agent/integrations/langsmith.py`
 
 ## 2. Model
 
-The model is configured in the `get_agent()` function in `agent/server.py`. By default it uses `anthropic:claude-opus-4-6`, but you can override it with the `LLM_MODEL_ID` environment variable:
+The model is configured in the `get_agent()` function in `agent/server.py`. By default it uses `openai:gpt-5.5` with medium reasoning effort, but you can override the model with the `LLM_MODEL_ID` environment variable:
 
 ```bash
 # Set the model via environment variable (uses provider:model format)
 LLM_MODEL_ID="anthropic:claude-sonnet-4-6"
 ```
 
-If `LLM_MODEL_ID` is not set, the default model (`anthropic:claude-opus-4-6`) is used.
+If `LLM_MODEL_ID` is not set, the default model (`openai:gpt-5.5`) is used.
+
+`max_tokens` is a maximum completion/output token budget, not the model's total context window. For OpenAI reasoning models, this budget can include both internal reasoning tokens and final response tokens.
 
 ### Switching models
 
@@ -137,7 +144,7 @@ Use the `provider:model` format:
 model=make_model("anthropic:claude-sonnet-4-6", temperature=0, max_tokens=16_000)
 
 # OpenAI (uses Responses API by default)
-model=make_model("openai:gpt-4o", temperature=0, max_tokens=16_000)
+model=make_model("openai:gpt-5.5", max_tokens=128_000, reasoning={"effort": "medium"})
 
 # Google
 model=make_model("google_genai:gemini-2.5-pro", temperature=0, max_tokens=16_000)
@@ -169,7 +176,7 @@ async def get_agent(config: RunnableConfig) -> Pregel:
         model = make_model("anthropic:claude-sonnet-4-6", temperature=0, max_tokens=16_000)
     else:
         # Full model for code changes from Linear
-        model = make_model("anthropic:claude-opus-4-6", temperature=0, max_tokens=20_000)
+        model = make_model("openai:gpt-5.5", max_tokens=128_000, reasoning={"effort": "medium"})
     
     return create_deep_agent(model=model, ...)
 ```

README.md

@@ -41,7 +41,7 @@ Rather than forking an existing agent or building from scratch, Open SWE **compo
 
 ```python
 create_deep_agent(
-    model="anthropic:claude-opus-4-6",
+    model="openai:gpt-5.5",
     system_prompt=construct_system_prompt(...),
     tools=[http_request, fetch_url, list_repos, get_branch_name, commit_and_open_pr, linear_comment, slack_thread_reply],
     backend=sandbox_backend,

agent/server.py

@@ -59,7 +59,7 @@ from .tools import (
 )
 from .utils.auth import resolve_github_token
 from .utils.github_app import get_github_app_installation_token
-from .utils.model import make_model
+from .utils.model import ModelKwargs, OpenAIReasoning, make_model
 from .utils.sandbox import create_sandbox
 from .utils.sandbox_paths import aresolve_sandbox_work_dir
 
@@ -183,8 +183,10 @@ def graph_loaded_for_execution(config: RunnableConfig) -> bool:
     )
 
 
-DEFAULT_LLM_MODEL_ID = "anthropic:claude-opus-4-6"
-DEFAULT_RECURSION_LIMIT = 1_000
+DEFAULT_LLM_MODEL_ID = "openai:gpt-5.5"
+DEFAULT_LLM_REASONING: OpenAIReasoning = {"effort": "medium"}
+DEFAULT_LLM_MAX_TOKENS = 64_000
+DEFAULT_RECURSION_LIMIT = 9_999
 
 
 async def get_agent(config: RunnableConfig) -> Pregel:
@@ -273,12 +275,14 @@ async def get_agent(config: RunnableConfig) -> Pregel:
 
     work_dir = await aresolve_sandbox_work_dir(sandbox_backend)
 
+    model_id = os.environ.get("LLM_MODEL_ID", DEFAULT_LLM_MODEL_ID)
+    model_kwargs: ModelKwargs = {"max_tokens": DEFAULT_LLM_MAX_TOKENS}
+    if model_id == DEFAULT_LLM_MODEL_ID:
+        model_kwargs["reasoning"] = DEFAULT_LLM_REASONING
+
     logger.info("Returning agent with sandbox for thread %s", thread_id)
     return create_deep_agent(
-        model=make_model(
-            os.environ.get("LLM_MODEL_ID", DEFAULT_LLM_MODEL_ID),
-            max_tokens=20_000,
-        ),
+        model=make_model(model_id, **model_kwargs),
         system_prompt=construct_system_prompt(
             working_dir=work_dir,
             linear_project_id=linear_project_id,

agent/utils/model.py

@@ -1,10 +1,25 @@
+from typing import Literal, TypedDict, Unpack
+
 from langchain.chat_models import init_chat_model
 
 OPENAI_RESPONSES_WS_BASE_URL = "wss://api.openai.com/v1"
 
 
-def make_model(model_id: str, **kwargs: dict):
-    model_kwargs = kwargs.copy()
+OpenAIReasoningEffort = Literal["none", "low", "medium", "high", "xhigh"]
+
+
+class OpenAIReasoning(TypedDict, total=False):
+    effort: OpenAIReasoningEffort
+
+
+class ModelKwargs(TypedDict, total=False):
+    max_tokens: int | None
+    reasoning: OpenAIReasoning | None
+    temperature: float | None
+
+
+def make_model(model_id: str, **kwargs: Unpack[ModelKwargs]):
+    model_kwargs: dict[str, object] = kwargs.copy()
 
     if model_id.startswith("openai:"):
         model_kwargs["base_url"] = OPENAI_RESPONSES_WS_BASE_URL